Natural disaster events that made headlines in 2020 and are projected to resurge in 2021 and throughout this decade
What it does:
Part I: "Scan and call":
- The user can upload a camera photo to the web and gets a prediction on the structural integrity of a building, which may fall into 5 categories: safe, less safe, risky, very risky, collapsed.
- (Was planned but not completed in time) If in danger predicted by the web application, people can use a hotline button to call several nonprofit and NGO emergency services.
Part II: "Data speaks"
- The user can upload the geographical information about an area of interest, including country, specific religion, longitude, latitude, and the application will predict the Community Preparedness Index for this area
How we built it:
- Part I: front end uses react and node.js and is edited in replit, backend is supported by Google Teachable Machine that can classify images
- Part II: the model is created from MLRun. We install MLRun on a Kubernetes Cluster by configuring Docker desktop on Mac, then prepare a dataset formatted similarly to the MLRun tutorial for ML training
Note: part I demo is in the presentation below, and part II demo is in the Youtube video. Thanks!
Challenges we ran into:
- Research-wise: narrowing down the problems underserved communities face during natural disasters and choosing what solution we wanted to provide
- Technology-wise: teachable machine model configuration; docker setup; getting permission to create and store the MLRun functions with a different name on the central database
Accomplishments that we're proud of:
- Understand many people will help themselves in terms of resources and location if we give them the tools to determine how safe they are at any particular building or location
- Understand adequate natural disaster relief for the poor doesn't have to be a greater burden on government or private organizations
What's next for Leviosa:
- Partner with mobile companies such as Android and T-Mobile to make the app a part of the OS update system
- Construct a better model using MLRun and hopefully conduct an image segmentation workflow using MLRun